A forecaster who learnt from a collections floor

In the winter of 2013, at a regional bank in Brisbane, I watched a collections model I had built do exactly what it was designed to do. It ranked overdue credit-card customers by the likelihood that a phone call would produce a payment, and the call centre worked down the list. The model was accurate. It was also, I realised when I read a sample of call notes, sending most of our attention to people who could pay if pressed, while customers who had quietly stopped coping sat at the bottom of the list until their accounts were written off. Nothing in the model was wrong except its question. I have spent the years since trying to ask better ones.

I grew up in Ballarat, where my mother kept the books for a stock-and-station agency and my father read the rain gauge every morning before work. I studied statistics and atmospheric science at Merrindale University in Melbourne, graduating in 1998, and joined Fairlie Weather Analytics, a small forecasting consultancy whose clients included grape growers, an alpine resort and two electricity retailers. The retailers paid for temperature forecasts because the newly opened wholesale market had made the cost of a wrong guess about demand immediate. I wrote my first demand model for one of them in 2000, and I have never quite left the subject.

In 2002 I moved to Brisbane to join the forecasting team at Kowrie Generation Corporation, a Queensland government-owned generator with gas-fired and hydro stations. My job was to tell the bidding desk how much the state would use tomorrow and in the next half-hour. I learnt that a forecast is a commercial instrument, judged less by its average error than by what it costs on the days it is wrong, and that the people who carry that cost deserve to see how it was made. I completed a master's degree in applied statistics at the Halloway Institute of Technology in the evenings, finishing in 2005.

Then I left energy for ten years, which I now count as the most useful decision of my career. At Orrinvale Communications in Sydney, from 2007, I built churn and pricing models for 2.4 million mobile and broadband customers. I learnt how a business decides which customers are worth keeping, and how retention discounts can quietly reward the restless at the expense of the loyal. At Coorabin Bank, back in Brisbane from 2012, I led credit and collections analytics, where the episode I began with took place. We rebuilt the collections strategy so that signs of hardship moved a customer towards assistance rather than further down a calling queue, and the share of hardship arrangements that customers completed rose from 38 to 57 per cent.

Wirrandel Energy recruited me in 2017 to lead customer and market analytics. Victoria's smart meters meant we already held interval data for most of our customers there, but we used it mainly to bill. Over four years my team built the foundations: a meter-data platform, pricing analytics for the default market offers introduced in 2019, the first analytics for our virtual power plant and an honest picture of which customers were leaving and why. I became chief data officer in March 2021, reporting to the managing director, with responsibility for data engineering, analytics, data science, forecasting and data governance across the company.

Most of what I have done since rests on one conviction from the bank: a data function should be judged by the decisions it changes, including the decisions it declines to support. We built an early-support model that flags households drifting into difficulty and sends its output only to the team that offers help. We rebuilt our rooftop-solar forecast from our own customers' meters after a spring in which our hedge book assumed demand that never arrived. We withdrew thousands of batteries from frequency markets until we could prove what they had done. And we built an emissions ledger with the same controls as our revenue, because our directors now sign both numbers.

I lead 260 people across Melbourne and Adelaide. I ask them to show their working, to bring me the cases a model gets wrong as well as its averages, and to describe every dataset by whom it concerns before describing what it contains. I attend the Audit and Risk Committee for privacy, data controls and the climate statement, and I present our data strategy to the full board twice a year. Directors rarely need more charts, I have found. They need the single example that tells them whether a model behaves as we claim.

Inside the industry I am probably known for three positions that have cost us money in the short run: that hardship signals must be walled off from collections, that a retailer should forecast its own customers' roofs rather than buy a regional average, and that an emissions figure deserves revenue-grade controls. Since 2022 I have also served on the board of Quorrell Street Financial Counselling, a not-for-profit service in Melbourne's western suburbs, where I see the other side of the bills we send. I am now preparing for one non-executive role on a listed or state-owned board.

Open to Board appointments · Speaking invitations

Languages / English